A Syntax-aware Multi-task Learning Framework for Chinese Semantic Role Labeling

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 5382–5392, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 5382 A Syntax-aware Multi-task Learning Framework for Chinese Semantic Role Labeling Qingrong Xia, Zhenghua Li∗, Min Zhang Institute of Artificial Intelligence, School of Computer Science and Technology, Soochow University, China kirosummer.nlp@gmail.com, {zhli13, minzhang}@suda.edu.cn Abstract Semantic role labeling (SRL) aims to identify the predicate-argument structure of a sentence. Inspired by the strong correlation between syntax and semantics, previous works pay much attention to improve SRL performance on exploiting syntactic knowledge, achieving significant results. Pipeline methods based on automatic syntactic trees and multi-task learn- ing (MTL) approaches using standard syntac- tic trees are two common research orienta- tions. In this paper, we adopt a simple uni- fied span-based model for both span-based and word-based Chinese SRL as a strong baseline. Besides, we present a MTL framework that includes the basic SRL module and a depen- dency parser module. Different from the com- monly used hard parameter sharing strategy in MTL, the main idea is to extract implicit syntactic representations from the dependency parser as external inputs for the basic SRL model. Experiments on the benchmarks of Chinese Proposition Bank 1.0 and CoNLL- 2009 Chinese datasets show that our proposed framework can effectively improve the perfor- mance over the strong baselines. With the ex- ternal BERT representations, our framework achieves new state-of-the-art 87.54 and 88.5 F1 scores on the two test data of the two bench- marks, respectively. In-depth analysis are con- ducted to gain more insights on the proposed framework and the effectiveness of syntax. 1 Introduction Semantic role labeling (SRL) is a fundamental and important task in natural language processing (NLP), which aims to identify the semantic struc- ture (Who did what to whom, when and where, etc.) of each given predicate in a sentence. Seman- tic knowledge has been widely exploited in many down-stream NLP tasks, such as information ex- ∗Corresponding author. 昨天 Yesterday ， , 汤姆 Tom 买 [Root] bought 了 一双 a pair of 鞋 shoes 。 . advpunc subj adjct obj punc adv root AM-TMP A0 V A1 AM-TMP A0 买.01 A1 Figure 1: Example of span-based (blue blocks) and word-based (red blocks) SRL formulations in a sen- tence, where the top part is its dependency tree. traction (Bastianelli et al., 2013), machine transla- tion (Liu and Gildea, 2010; Gao and Vogel, 2011) and question answering (Shen and Lapata, 2007; Wang et al., 2015a). There are two formulations of SRL in the com- munity according to the definition of semantic roles. The first is called span-based SRL, which employs a continuous word span as a semantic role and follows the manual annotations in the Prop- Bank (Palmer et al., 2005) and NomBank (Mey- ers et al., 2004). The second is word-based SRL (Surdeanu et al., 2008), also called dependency- based SRL, whose semantic role is usually syn- tactic or semantic head word …

Content truncated to 500 words